some question about zero-shot

[younginger]

"I found that the test results of my model after multiple training sessions were the same as the zero-shot results. I checked the intermediate trained model parameters and discovered that their test results, mAP and AR, were exactly the same. Why is this happening?"